Picture the scene. You've not been feeling well for some time, so you go to your GP. Your unexplained weight gain is put down to diet. Your high blood pressure is now being treated by two medications, and although we're treating each symptom, it's not all adding up, something still feels fundamentally wrong. The bigger picture is being missed.
One in 17 people in the UK has a rare disease. And often, when we look at diseases such as Cushing's or Conn's syndrome, the individual symptoms could be any number of things. But it's the combination that adds up to a diagnosis. Which in turn means getting diagnosed with Cushing's syndrome can become a medical marathon, seeing seven or eight clinicians on average before being able to get the right help.
OpalMedica is a Manchester-based medtech startup, founded by Sara Elgott, who spent over thirty years working in rare diseases across the pharmaceutical industry, and David McMinn, a PhD researcher and medical writer. Their mission: speed up detection so that people can get the right help, sooner. And they came to Fuzzy Labs to build it.
The challenge
Most patients wait years before getting a diagnosis. OpalMedica set out to change that with a system that reads what a GP writes during a consultation and raises a flag to whether this combination of symptoms could be pointing to either Cushings or Conns syndrome.
That's easier said than done. Open a GP's consultation notes and you'll find a completely different language from the formal medical record. Shorthand such as "Bp++" or "Low k" are common and don't map to the NHS's standardised codes. OpalMedica's idea was to work on the actual notes instead: the free text a doctor types during the consultation, helping GPs identify patterns in patient symptoms that may indicate a rare disease and support earlier referral to specialist care, right there in the room.
What we built
We built OpalMedica a NLP/symptom-extraction pipeline: it pulls symptoms out of a GP's free-text notes, matches them against a database of known symptoms, and scores the combination to flag when a rare disease is worth a closer look.
Reading what doctors actually write
The first stage is symptom extraction. A clinician enters consultation notes in whatever informal shorthand they normally use, and a small, fine-tuned language model identifies anything that looks like a symptom.
The team made a deliberate decision not to reach for an off-the-shelf, general-purpose model here. In a medical setting, a clinician needs to understand exactly why a prediction was made: which symptoms, how they were weighted, what each one contributed.
Understanding what's not there
Context matters as much as content. A note saying "no stretch marks" means something very different to "stretch marks noted", even though the words are nearly identical. The pipeline was built to catch this kind of negation, so a symptom that's been ruled out isn't mistaken for one that's present.
Matching symptoms to what's known
Each extracted symptom gets compared against a hand-built database of known symptoms for Cushing's and Conn's. Then every symptom was expanded: hypertension sits in the system alongside "high bp", "bp++", "HT - Hypertension", "high blood pressure disorder" and more.
A language model then handles the matching, finding the closest known symptom for whatever term comes through.
Scoring the risk
Each matched symptom carries a weight, set by OpalMedica's clinicians based on how strongly it indicates each condition. The total is then compared against defined thresholds to produce one of three outputs: unlikely, possible, or likely.
How we built it
The project ran in two two-week sprints. Oscar, one of our top engineers, led the technical build throughout, working closely with the OpalMedica team, including Dr Safwaan Adam, Consultant Endocrinologist at The Christie NHS Foundation Trust in Manchester and clinical consultant to OpalMedica. The OpalMedica team supplied the clinical knowledge that shaped the tool: symptom weights, diagnostic thresholds, and a precise picture of how it needed to behave in a real surgery, with Dr Safwaan's specialist expertise informing the clinical decision logic underpinning the models.
Oscar understood the clinical reasoning well enough to anticipate edge cases, spot gaps in the data, and bring forward additional solutions. The whole process was collaborative, and by the end of the project, Oscar knew enough about rare disease presentation that he very nearly beat the model in a head-to-head test. Very nearly.
About a third of the way into the first sprint, the scope did change. Cushing's is so rare we would need 10 million or more patient records just to find enough cases to validate the model properly. OpalMedica added Primary Aldosteronism (”Conn's syndrome”) mid-build. It's a different condition, but the pipeline requirements are similar, and there was more data available to validate against.
The demo interface was designed to feel like a GP's existing desktop. A clinician types their consultation notes and gets a prediction with the reasoning shown alongside: which symptoms were matched, how they were weighted, and why.
By handover, we were able to deliver a working prototype, a private API, an NHS-inspired demo interface, evaluation reports for both conditions, and a full technical roadmap for production.
The results
Of the 80 Conn's patients in OpalMedica's survey data, 74 had needed to see more than one healthcare professional before getting the right diagnosis: a 92.5% miss rate on the first consultation.
Fed the same patients' symptoms through that process, the pipeline flagged 51 of them, two-thirds, as possible or likely.
What's now possible
The original grant funded four weeks of engineering, which was enough to show the concept works. What OpalMedica is building now is something considerably bigger: a pharma contract in progress, a pilot across Greater Manchester approved, working with multiple NHS and academic partners to support future evaluation.
We were glad to be a small part of the story, and can’t wait to see all OpalMedica achieve as quick diagnosis becomes less rare, helping countless people receive the right diagnosis sooner.
"Working with Fuzzy Labs on the development of our MVP for OpalMedica Clinical Flags has been a genuinely brilliant experience. From the very beginning, the whole team understood not just the technical challenge, but the wider mission behind what we are trying to achieve for people living with rare diseases.
Their approach has felt truly aligned with our values around building technology responsibly and to really make a difference to people's lives.
Our main contacts were Robbie and James who looked after us from the start, but a special mention to Oscar, who consistently exceeded expectations throughout the project. His ability to translate complex ideas into something practical and scalable has been exceptional.
— Sara Elgott, Co-founder, OpalMedica
.png)
